- role
- applied AI engineer · LaunchGood
- location
- Ontario, Canada
- status
- shipping · MEng
~/about
Suleyman Kiani
Applied AI Engineer · LaunchGood
I build AI agents and ship them with their evals attached. I am an Applied AI Engineer at LaunchGood, which I joined in October 2026 after a year in equipment finance at Mitsubishi HC Capital. A personal multi-agent OS runs the rest.
Since October 2026 I have been a senior Applied AI Engineer at LaunchGood, in the pod that owns AI initiatives. Before that I spent a year funding equipment-finance deals at Mitsubishi HC Capital, over quota every month from my second, and built the desk a quoting tool that corporate IT now owns and is deploying. Alongside that I run two production products with paying users.
My MEng in Computing & Software at McMaster is research complete: Dr. William Farmer and Dr. Richard Paige reviewed the running system in September 2026 and accepted the project. I am writing the final report now, and the degree is expected in 2027. The project is Mac Study Companion, a study tool for neurodivergent learners, described below. A+ in simple type theory, A in microservice architectures.
Off the clock I built my own JARVIS: a personal multi-agent operating system I run across my own machines. Scheduled jobs and email-triggered workers carry out agent tasks on a private mesh, each one sandboxed behind a fail-closed egress allowlist so an untrusted inbound document never reaches a credential. Work that dies mid-flight is re-driven a bounded number of times and then parked for a human rather than retried forever, a dead-man switch watches the health monitor itself, and a local model answers over my own notes with citations. The engineering and the finance feed each other.
mac-study-companion
github.com/kianis4M.Eng research: study notes that cannot make things up.
A virtual TA built for neurodivergent learners, designed first for students with ADHD. Lecture recordings become study notes where every released sentence cites the minute of audio it came from, and any sentence the checker cannot prove is held back in place with its reason shown. A fail-closed release gate discharges the verification conditions in Z3; the pipeline state machine is model-checked in TLA+; the Socratic tutor is given only the steps the student has demonstrated, so it cannot reveal an answer it was never told. Nine FastAPI services over a Redis Streams bus. Accepted by both supervisors in September 2026.
mike-ross-ai
mike-ross.caAgentic legal RAG: cited Canadian-law answers.
A LangGraph state machine runs hybrid vector retrieval over federal and provincial statutes (MongoDB Atlas) with a Gemini 2.0 Flash reasoning core, returning schema-validated, section-cited answers that block hallucination. FastAPI + Next.js, streamed over SSE. Built for Delta Hacks 12.
incite
skomp.studioWhite-label studio platform running a real business.
Booking, waitlists, memberships and Square payments for studios that ran on paper. Tenant isolation enforced in the data layer by a scoped Prisma client, 74 data models across 196 API routes, and 8 required CI checks on every merge, including more than 16,000 Vitest cases and two Playwright suites. Its launch client, Solstice Pilates, runs its entire operation on it.
applify-ai
applify-ai.comResume-tailoring ML product.
Event-driven LLM pipelines generate ATS-optimized, job-targeted resumes from any job description. ~150 paying users.
skompxcel
skompxcel.comMentorship platform for CS students.
Algorithms, systems design, mock interviews, resume reviews. 100+ learners.
chess 1650 rapid · kickboxing